Find and Select the Right File Storage for your Applications. Philippe Nicolas, KerStor

Transcription

1 Find and Select the Right File Storage for your Applications Philippe Nicolas, KerStor

2 SNIA Legal Notice The material contained in this tutorial is copyrighted by the SNIA. Member companies and individual members may use this material in presentations and literature under the following conditions: Any slide or slides used must be reproduced in their entirety without modification The SNIA must be acknowledged as the source of any material used in the body of any document containing material from these presentations. This presentation is a project of the SNIA Education Committee. Neither the author nor the presenter is an attorney and nothing in this presentation is intended to be, or should be construed as legal advice or an opinion of counsel. If you need legal advice or a legal opinion please contact your attorney. The information presented herein represents the author's personal opinion and current understanding of the relevant issues involved. The author, the presenter, and the SNIA do not assume any responsibility or liability for damages arising out of any reliance on or use of this information. NO WARRANTIES, EXPRESS OR IMPLIED. USE AT YOUR OWN RISK. 2

3 Abstract Title: Abstract: Many businesses are linked to file storage technologies as many of these new, recent and even existing applications morph now, rely and support file based data. At the same time, the volume of data explodes especially the file data type which now represents by far the larger portion of enterprise data. With the complexity and variety of market solutions, the challenge for IT buyers and storage managers is to choose and adopt the most adapted solutions aligned to their business and IT needs to address their current and future challenges with a special attention to compliance and data retention regulations, Backup and Archiving, ILM and Tiering. This session covers the most common deployed applications, their attributes in term of file storage needs and maps these to file storage solutions available in the industry with technologies details and advantages. Various technologies are presented in this session, among them: Clustered, SAN-based, Distributed, Parallel File Storage and the very last approach Cloud Storage. Learning objectives: 1. With a top-down approach, this tutorial improves file storage technologies positioning and understanding aligned to applications needs and challenges. 2. The presented survey, technologies segmentation and features matrix, helps end-users, IT and file storage buyers to select, choose and adopt the right solution. 3. And finally this content contributes to the promotion of file storage technologies in an agnostic philosophy. Audience: IT & Storage Architect, IT Manager and Buyers. 3

4 Agenda Applications Convention, Needs & Challenges Type, Characteristics, I/O Patterns & Access The Right File Storage for Each Application Application attributes and File Storage details Basic & Advanced File Services Conclusion 4

5 Applications Convention, Needs & Challenges Type, Characteristics, I/O Patterns & Access

6 Applications Convention No distinction between software services above File System/Storage logic & layer Everything is an Application: Database, Web Server, Computing (HPC), Backup, Archival & ILM, Video streaming even multi-tiers or multi-layers CRM DB File Storage or Video Streaming File Storage Remark - Exclusion: Application can run on many various local disk file systems in a single server-storage domain but this classical approach is not covered in this tutorial Application File Storage 6

7 End-User Needs and Examples of use End User Needs Better scalability Capacity: fast growth of volume of data, # files, filesize Performance: IOPS, BW, frame/sec Data Sharing: avoid data duplication, aggregate more servers Larger server can be expen$$ive More Availability, no Downtime Local (clustering, failover ) Remote (wide failover + data replication ) Global (multi-sites) Easy Manageability and Administration Migration and Consolidation (data movement) Remote site and file server management Industry Standard Protocols, components, COTS Advanced features Load Balancing, Quotas, Security, Data Protection (snapshot, replication ), ILM & Classification, Data Reduction, Encryption, Content Indexing & Search, Reporting & Statistics, XAM Cost Reduction $/TB, $/IOPS, $/BW, $/Transac fs op., $/NFSops Examples of use High Availability Clusters (local & geographic) Multiple point of presence Scaling applications Web Servers - Read mostly/load balanced Databases/OLTP/DW - Mostly use direct I/O Distributed and Parallel app. and fast failover Data acquisition and «demanding» computing Systems, App. and Data Consolidation/Migration, ILM Tech. Refresh Obsolescence Off-host processing Based on shared file system Can also use by Point-in-Time copy techniques (not related to our data sharing definition) Cloud Storage Online Backup, Data Archiving, Capture and Indexing, DR and new tier of storage consideration 7

8 Evolution of File Services Data Volumes & Storage Utilization Towards a commodity File Service Original File Servers (1st gen.: ftp) Traditional File Servers (2nd gen. = gen 1 + nfs & cifs) Extended File Servers (3rd gen. = gen. 2 + WebDAV + http) New Generation File Services vertical + collaboration + massive (4th gen. = gen. 3 + DB back-end + Metadata structure + Asym. & Massive) Budget & Time Constraints Average disk allocation for Windows systems: 25-40% for Unix/Linux systems: 30-45% for iseries and z/os : 60-80% File Server/NAS considered as NAS Target for Backup (with DeDup) & Archiving (WORM CAS) for Virtual Servers/Machines images for Vertical, Collaboration, HPC, Video, Cloud 8

9 Applications & File Storage CADCAM HPC File Sharing Multimedia Remote File Sharing Web Server DB ftp Data Acquisition Video Streaming Archiving ECM Data Analysis CAS* Parallel FS CFS* NAS SaaS WAFS* SANFS* NAS Cluster Distributed FS Massively Scalable FS Cloud Based on Applications characteristics and I/O behaviors, the goal is to select the right File Storage technologies among many, many existing approaches * CFS: Cluster File System SANFS: SAN File System WAFS: Wide-Area File Service (WAN Optimization & Acceleration) CAS: Content Addressable Storage 9

10 Applications characteristics Workload profile OLTP Small Data Mart Home Directory Large Scale Streaming (web farm) High-Frequency Meta-Data update (small file create/delete) Latency sensitive High Med Low Low High Throughput High R/W High read Low High read High write Concurrent sharing Caching (re-read rate) High High Low High read Low High High High Low Low 10

11 File Storage usage Processor Farms Office Env. Archive Data Streaming Media Database Web and Content Management Typical applications Financial simulation, Grid computing, Asic Simulation, Oil & Gas Applications, Rich Digital Content creation Rendering Spreadsheet, Word processing, Presentation, Pictures editing Medical records & imaging, insurance policy stores, gov. records, check storage applications, video surveillance Online content delivery Direct Pay per View News distribution online radio audio streaming social networking Exchange, SQLServer, Oracle, other OLTP Web farms IIS, Apache, CAD, SW dev. Usage of File Storage Used as a Virtual memory Used as a shared storage pool Used as long retention storage for online archive of records, documents and images Used as a large and performance storage pool Used as alternative to traditional raw and local disk file system Used as a generic repository I/O patterns & access Equal mix of reads and writes Sequential and large transfer sizes Random reads and small request sizes Large sequential access Mixed small and large sequential transfer Mix read/write small size Small random reads 11

12 The Right File Storage for Each Application Application attributes and File Storage details Basic & Advanced File Services

13 Some approaches Top-Down Understand, monitor and profile application I/O pattern Configure network stack, volumes, LUN, stripe size and file system (block size ) Tune, verify & control, adapt Bottom-Up Configure network stack, storage, LUN and volumes, stripes + file system Build and align application I/O size to the above one (need source) Tune, verify & control, adapt Other Too many various I/O access and behaviors Storage and Application modification not possible 13

14 Web Server farm Parallel DB Application Web Server farm with http, ftp services Characteristics: Thousands/millions concurrent IOPS read oriented, few updates, multiple data/file format, potential DB addition Needs: Medium to High Read throughput, performance scalability (linear) Options: Load balancer in front Parallel database Characteristics: Concurrent access (1 physical DB accessed by multiple instances), thousands/millions IOPS, latency influence Needs: High R/W throughput, performance scalability (linear), integrated failover Options: snapshot, HA/BC/DR Configuration Cluster File System From 2 to 16/32 nodes 14

15 Cluster File System Cluster File System (CFS), also named Shared Data Cluster A Cluster FS allows a FS and files to be shared Centralized (asymmetric) & Distributed (symmetric) implementation Centralized uses master node for meta-data updates, logging, locking All nodes understand Physical (on-disk) FS structure The FS is mounted by all the nodes Web Web Server Server Single FS Image (Cache Coherence) Lock Mechanism Distributed or Global Lock Management (DLM/GLM) //DB //DB //DB /DB Web Server Cluster File System SAN Web Server 15

16 Multimedia application Application Audio/Video Streaming, Clip Editing, Movie rendering Characteristics: Hundreds/thousands concurrent sequential BW operations (acquisition, edition ), multiple data/file format Needs: Medium to High RW throughput, performance scalability (linear) for servers, hundreds of clients tolerance, thousands of application consumers, open protocols, heterogeneous OS Options: Tiering, Classification, Capacity Optimization, Indexing Configuration SAN File System 16

17 SAN File System SAN File System (SAN FS) aka SAN File Sharing System A SAN FS allows files to be shared Client/Server or Master/Slave (aka asymmetric) model Mixed role between direct data access with host based thin software and NAS access Flexibility of network FS at SAN speed Designed to support hundreds or thousands of nodes Lock Mechanism & Cache Coherency Meta-Data Server (NAS Server) App. App. App. SAN Client SW Workstation 17

18 CFS vs. SAN FS Characteristics & Features Cluster FS SAN FS Shared elements File System and files Files only # of nodes Dozens Hundreds Heterogeneous OS No Yes Tolerance of Distance (between server and clients) Dedicated Meta-Data Server(s)* required Physical File System layout knowledge Single File System Image (SFSI)/Cache coherency File Sharing behavior Limited No (except centralized design or configuration) All nodes (Cluster FS currently requires same OS) Yes Exportable by all nodes in the cluster Important Yes, usually Meta-data server only (clients may understand if same OS) No No file sharing by clients File sharing by meta-data server especially if NFS/CIFS based * Meta-Data Server aka Master Server 18

19 Data Acquisition, File Sharing Application Data Acquisition Characteristics: Hundreds/thousands concurrent sequential BW operations, multiple data/file format Needs: Medium to High RW throughput, performance scalability (linear) for servers, hundreds of clients tolerance, open protocols File Serving/Sharing, File Storage Consolidation, Data/File repository, Office application even remote file access Characteristics: Hundreds/thousands concurrent IOPS, multiple data/file format Needs: Performance scalability, hundreds/thousands of clients/applications tolerance, embedded data protection, directory integration, open protocols DataBase on NFS Characteristics: Hundreds/thousands concurrent IOPS Needs: Performance (DIO, AIO ) + scalability, optimized failover, embedded data protection, open protocols Configuration Distributed approach with Network File System, aggregation of file servers (FAN, NFV/NFM) and WAFS/WAAS/WADS with SMB2 19

20 Network File Server aka NAS Distributed File System General Characteristics Network transparency, User Mobility, Fault Tolerance, Scalability, File Mobility No File Server aggregation by default NFS primarily for Unix and CIFS for Windows (NAS protocols) Asymmetric (Client/Server) architecture Uses TCP/IP (UDP for NFS in the past, NFS over RDMA) De facto standards today Too chatty /verbose for remotes file access NAS Server SAN Client SW Workstation Evolution (recent NAS protocols development) SMB/CIFS v1 to v2 «WAN optimized» (v 2.002): >30 times faster compared to SMB1 over WAN and 2-10x on LAN, compound mechanism (aggregation of multiple requests only 19 commands, reduce round trips), durable file handles, larger buffer sizes,, sym. links, secure and robust NFS v2 to 3 and now v4 (v4.1): Only 1 port (2049), stateful, compound operations, client caching + delegation, security (authentication + Windows ACLs), migr. + repl., Unix/Linux & Window support, RDMA, TCP (only), Namespace ext. (Mirror mounts & referral) + FedFS + pnfs (v4.1) 20

21 Parallel NFS with NFSv4.1 (pnfs) pnfs is about scaling NFS and address file server bottleneck Same philosophy as SAN FS (master/slave asymmetric philosophy) and data access in parallel Allow NFSv4.1 client to bypass NFS server No application changes, similar management model pnfs extensions to NFSv4 communicate data location to clients Clients access data via Fibre Channel, FCoE & iscsi (block), OSD (object) or NFS (file) NFSv4.1 Client(s) pnfs protocol NFS Server Control protocol MetaData Server Storage-access protocol Data/Storage Servers 21

22 FAN = File Area Network Definition (SNIA Dictionnary) A namespace-based network-oriented infrastructure for files that includes a decoupling layer which separates logical file access from physical file location. This decoupling layer enables a variety of services (e.g., replication and migration) to be applied to files and filesystems Applications and Users Storage Consumers Logical Access Decoupling control Services Storage Physical Access NAS Storage Resources 22

23 Namespace concept Clients or Consumers Namespace Aggregation CIFS/NFS IP/Ethernet File Servers or Producers Shared Namespace Proprietary approach internal aggregation of same brand/model file/storage servers Global Namespace Open approach external aggregation of individual file servers + shared namespace if any 23

24 Decoupling Approaches Client-based (out-band) Hybrid (dual-band) Network-based (in-band) Clients Tree-level Decoupling Tree Tree Tree File-level Decoupling File File Storage 24

25 File Virtualization & Management File Virtualization Capacity to mask physical location across (file) servers and provide logical access among them Seen as one logical entity No network, NAS protocols or client (consumer) related Example : Cluster File System Network File Virtualization (NFV) Integration of network, NAS protocols and clients (consumers) on top of (file) servers (notion of FAN) Example : DFS, Automount, NFS v4 namespace extensions (mirror mounts, referral & FedFS) File Management = File Virtualization + File Services File Services: Replication, Failover, Load Balancing, Quotas, Security, Data Protection (snapshot, replication ), ILM & Classification, Data Reduction, Encryption, Content Indexing & Search, Reporting & Statistics, XAM Network File Management (NFM) = NFV + File Services 25

26 Data intensive & HPC application Application Data intensive application, Data Collection/Logging, Data Analysis, High Performance Computing, High stress application Characteristics: Hundreds/thousands concurrent BW & IOPS, large data transfer, multiple data/file format Needs: Very High RW throughput, logical consolidation, performance scalability (linear) for servers (1000s), 100s to 10000s of clients tolerance, open protocols, PBs of capacity, full embedded resiliency and redundancy, global, permanent and ubiquity access and presence, notion of global/shared namespace, standard or proprietary file access, security and privacy Options: Load balancing, Tiering, Storage Capacity Optimization Configuration Distributed File System (back-end data distribution and parallel) 26

27 Distributed implementations Aggregation Union of storage servers Symmetric, Asymmetric or P2P Central Authority aka Master node or Masterless Parallel vs. non-parallel File striped and concurrent access, single node access Shared-nothing model Aggregation of file storage/servers User-mode vs. Kernel/System-mode File-based vs. object-based CIFS/NFS NAS Cluster WAFS NFM/NFV FAN Distributed Parallel Notion of Shared/Private Namespace (storage nodes wide) RAIN, Grid implementation with embedded data protection and resiliency No storage nodes are fully trusted and are expected to fail at any time Extended features Policies enforcement for Automation, Snapshot, CDP, Replication, Mirroring, ILM/FLM/Tiering, Migration, Load balancing pnfs PVFS Own or Proprietary 27

28 P2P Implementations Interesting implementation with aggregation of other machines storage space with or wo a central server (asym or sym P2P) Ex: P2P File Sharing System such as music, mp3 Napster Gnutella BitTorrent 28

29 Industry Implementations Symmetric Philosophy Aggregation of independent storage servers File entirely stored by 1 storage server (no file striping) Aggregation of homogeneous storage servers File is striped across storage server Shared Namespace Shared Namespace Data/File striping across servers 29

30 Google File System Internal deployment but used by WW end-users Proprietary approach Asymmetric philosophy Thousands of chunk servers with 64MB chunk size (stripe unite) research.google.com/pubs/papers. html GFS v2 Chunk size = 1MB!! Hundreds of Distributed Masters (100 millions files managed by 1 master) Source Google 30

31 Hadoop DFS Apache project Highly fault-tolerant built in Java Large data sets Asymmetric philosophy Files striped across DataNodes hadoop.apache.org 31

32 Other examples Lustre Open source object-based storage system (based on NASD study from Carnegie Mellon Univ.) Asymmetric Philosophy Notion of Object (OST/OSS) PVFS (Parallel Virtual File System) now in 2 nd gen. MogileFS Project from Clemson Univ. and Argonne National Lab. Open source and based on Linux Asymmetric philosophy Application-level Distributed FS (no kernel module and no POSIX compliant), Asymmetric Open-source and Local filesystem agnostic Fully redundant, automatic file replication Flat Namespace and Shared-Nothing Haystack, FineFS, again pnfs* and CAS* implementation * pnfs: Parallel NFS with NFS v4.1 CAS: Content Addressable Storage 32

33 Cloud Storage Cloud Storage is simply the delivery of virtualized storage on demand. SNIA proposes Data Storage as a Service (DaaS), which means delivery over a network of appropriately configured virtual storage and related data services, based on a request for a given service level. Cloud Storage Initiative ( Cloud Data Management Interface (CDMI) Cloud Storage Use Cases and Reference Model White Paper «Cloud Storage for Cloud Computing» 33

34 Cloud Storage Models Only Storage Cloud (cloud-attached storage) Application + Storage Same Cloud Application + Storage (primary Cloud) + Secondary Storage for Backup, Archiving or DR (Secondary Cloud) Application Cloud connected to Storage Cloud Notion of Private or Public Cloud 34

35 Cloud Storage Application Online & Primary File Storage, Online Backup and Data Archiving, DR A new tier of Storage Characteristics: Thousands/millions concurrent IOPS, multiple data/file format Needs: High resiliency and geo data distribution, security and data privacy, performance and capacity scalability, open data access protocols or open API, contract based (SaaS) Options: embedded data protection, data distributed across file/storage servers Configuration Distributed and Scalable File System, Cluster File System FUSE (Filesystem in USEr space) VLFS as the industry introduces VLDB File System -> File Manager -> File Storage Everything is a File 35

36 Basic & Advanced File Services Basic Global Namespace (File Virtualization) Organize storage in an overlay namespace Migration Move files from one server to another Tiering / ILM Move files via policy to the best storage Load Balancing Move files to better distribute capacity or load Data Protection Snapshot to support online data protection Replication as a BC strategy Reporting & Statistics Advanced HA/BC/DR* Associated with data replication Data Classification and Optimized Placement Data value and storage characteristics alignment Storage Capacity Optimization DeDuplication, Reduction and Compression Quota Management Report and enforcement Content Indexing & Search File content to optimize file placement Search for legal discovery Application Acceleration Local and Distributed file access Notion of Distributed ILM/Tiering Security Access Control, auditing and Encryption * High-Availability/Business Continuity/Disaster Recovery 36

37 Applications/File Storage Matrix HPC DB OLTP Web farm Office Env. Multimedia Video Data Acquisition Archiving Cloud Storage ILM Other Vertical CFS (x) x x x x x x x SANFS x (x) x x x NAS (+ FAN) x x x x x x x x x x NAS Cluster x x x x x x x Distributed (& Parallel) x x x x x x WAFS (WAN Opt. & Acc.) x x CAS/Worm (with NAS access) x x 37

38 Conclusion

39 Conclusion File System and File Storage technologies are key for current and next IT and Information, Data and Storage Challenges Scalability is a multi-dimension metric (Performance, Capacity, Availability and Manageability) Asymmetric (Master/Slave) seems to be a more scalable philosophy Unified, Global Namespace is fundamental File/Object approach is superior (pragmatic) on block approach Reliability & Security are key (authorization, authentication, privacy ) Think standard (de facto and industry) as many of them evolve (pnfs) Commodity hardware is more and more common (reality) Many approaches and philosophies in the industry There is no single solution that is superior in all cases BUT these approaches deliver real applications and business benefits for different applications needs Study and choose the one which delivers the best value for you 39

40 SNIA Tutorials references Please check also the following tutorials Check out SNIA Tutorial: The File Systems Evolution Check out SNIA Tutorial: Object-Based File Systems: an overview Check out SNIA Tutorial: Global Namespaces for Summer 40

41 Q&A / Feedback Please send any questions or comments on this presentation to SNIA trackfilemgmt@snia.org (File Systems & File Management) Philippe Nicolas Many thanks to the following individuals for their contributions to this tutorial. - SNIA Education Committee 41

42 Find and Select the Right File Storage for your Applications Philippe Nicolas, KerStor